Disable shared object versioning

[m0nsky]

Shared object versioning was added in llama.cpp on November 11, 2025 (see this PR).

Due to the way we distribute our binaries (no versioned filenames), this was causing issues on platforms that did not use glibc (musl), or the dependency loader (for example, applications distributed with the .so libs in the executable root).

m0nsky@m0nsky-g14:~/publish$ ldd libggml.so
        linux-vdso.so.1 (0x00007ffdfb93d000)
        libggml-cpu.so.0 => not found
        libggml-cuda.so.0 => not found
        libggml-base.so.0 => not found
        libstdc++.so.6 => /lib/x86_64-linux-gnu/libstdc++.so.6 (0x0000735f13e00000)
        libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x0000735f141c8000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x0000735f13a00000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x0000735f140dd000)
        /lib64/ld-linux-x86-64.so.2 (0x0000735f1420c000)

I've compared the possible approaches to counter this (patchelf, update the CMakeLists or symlinks), and I think this is the cleanest approach that will not break if llama.cpp starts distributing additional libs in the future.

This PR disables shared object versioning for the deps, I've ran an Update Binaries action over at my fork which succeeded, and confirmed this fixes the issue on Ubuntu 24.04.

Before:

m0nsky@m0nsky-g14:~/publish$ objdump -p libggml-cuda.so | grep SONAME
  SONAME               libggml-cuda.so.0

After:

m0nsky@m0nsky-g14:~/publish$ objdump -p libggml-cuda.so | grep SONAME
  SONAME               libggml-cuda.so